FR2477003A1 - APPARATUS FOR RECOGNIZING FORM OF ELECTRIC SIGNALS OF PHYSIOLOGICAL ORIGIN - Google Patents

Abstract

There is described a structure recognition device and a method which is capable of detecting the presence of ventricular fibrillation automatically in an electrocardiogram. For this purpose, the electrical signals received by the electrodes applied to the patient are gated at intervals and the magnitudes of the gatings are converted from the analogous form into the digital form. The gatings standardised to a predetermined maximum value are then analysed in order to detect parameters of the signal which are characteristic of ventricular fibrillation.

Description

 The present invention relates to a method and an apparatus for recognizing shapes of electrical signals of physiological origin. The invention can be applied to the automatic diagnosis of cardiac conditions, in particular for the detection of conditions which may require defibrillation.

 Ventricular fibrillation often occurs during a heart attack and is commonly fatal unless treatment is applied quickly. Ventricular fibrillation is a condition in which the heart has stopped pumping or is affected by irregular spasmodic contractions, accompanied by chaotic electrical activity.

 The activity of a patient's heart can be monitored using an electrocardiograph, which receives electrical signals from the heart through two electrodes, which are usually attached to the patient's chest, and which display the electrical impulses received on a screen, usually a cathode ray tube, using an appropriate time base. The trace obtained can be observed by a doctor and it indicates the behavior of the heart. However, a correct interpretation of an electrocardiograph (ECG) curve requires considerable training because of the diversity of cardiac conditions giving rise to different curves, some of which are superficially similar. An incorrect diagnosis from an EKG can have very serious consequences.

 It is desirable to process the output signals of an electrocardiograph automatically to distinguish the normal electrical behavior of the heart (normal sinus rhythm) from abnormal behavior, in particular to highlight a fibrillation which generally requires very rapid treatment. Such processing can be carried out using a series of electronic circuits which together constitute a pattern recognition system. However, such an arrangement has the disadvantage that the characteristics of the signal received from the patient, and which are demonstrated by the shape of the electrocardiogram when displayed on a screen, are likely to be distorted by electronic circuits, especially RC circuits, used, so that one can get an incorrect diagnosis of certain heart conditions.

This difficulty is accentuated by the fact that one cannot use a single measurement criterion to safely distinguish ventricular fibrillation from other cardiac conditions and it follows that several parameters have to be analyzed. This makes the necessary circuits very complicated. , bulky and expensive and this results in an increased likelihood of distortion.

 The present invention relates to a method and apparatus for analyzing electrical signals, where the signal distortion by the apparatus itself is minimized.

 The invention is applicable to an automatic analysis of physiological electrical activity, in particular the electrical behavior of the heart, in particular for distinguishing ventricular fibrillation from a normal sinus rhythm and other conditions, such as ventricular tachycardia.

 In accordance with one aspect of the present invention, there is provided a pattern recognition apparatus, comprising two electrodes arranged to be connected to the body of a patient so as to receive an electrical signal varying on either side of a line isoelectric, means for sampling the received signal at intervals and for converting the magnitudes of the samples from an analog form to a digital form, an automatic gain control device for calibrating the magnitudes of the digitally converted samples with respect to a predetermined maximum value and a computer arranged to analyze the calibrated samples to detect an abnormal physiological condition.

 Embodiments of the invention will be described below with reference to an electrocardiograph but it goes without saying that the invention, in its broadest aspect, can also be applied to the recognition of forms of electrical signals originating from other sources.

 In one embodiment, the apparatus according to the invention, when used as an electrocardiograph, comprises two electrodes which can be fixed on the body of a patient so as to detect the electrical activity associated with the beating of the heart. The electrodes can be implanted but it is generally more advantageous to use external electrodes which are fixed on the chest. The electrodes are connected by suitable wires to the rest of the device which processes the received signals.

 The signals received through the electrodes are applied to an analog-to-digital converter in which they are expressed in digital form for further analysis. Once expressed in digital form, the information obtained at the output of the electrocardiograph is not modified by the following circuits.

The analog-to-digital converter samples the signals received from the electrodes at suitable intervals, for example every 0.004 seconds, and the information thus obtained can be stored in a digital memory.

When analyzing an electrocardiograph signal, it is desirable to study the QRS complex of the obtained signal, which is the characteristic component of a normal sinus rhythm. However, the QRS complex is generally accompanied by other signal components, which can have a considerable amplitude, such as the components
P and T. To eliminate these components, it is possible to attenuate all the frequencies of the original signal below 3 Ht and above about 18 Hz, ie the frequency range in which the complex QRS is located.

This problem can be solved by inserting low-pass and high-pass filters into the circuit between the electrodes and the analog-to-digital converter. As a suitable filtering device, a bandwidth filter of the "Sallen and Key" type can be used. The RC elements involved in such a filter create a small deformation in the waveform received by the converter, but it has been found that this deformation was small enough not to seriously interfere with the subsequent analysis of the signal.

When the signals converted digitally in real time have been memorized, they are first subjected to an automatic gain control (AGC) subroutine in order to calibrate their amplitudes; this is necessary because the amplitude of an EKG signal can vary considerably from patient to patient. The subroutine AGC takes the maximum-value from an appropriate number of values of successive samples (for example 250) and it deduces from the value a scale or framing factor by which all the samples are then multiplied. This scaling factor is chosen so that the maximum value, after multiplication, corresponds to a maximum limit which is determined by the operator. Referring to FIG. 1, it can be seen that the trace or curve represented in good sense corresponds to an electrocardiograph curve which is not multiplied, the maximum limit established by the operator being indicated in the figure; fig. 2 shows the appearance of the curve after multiplying all the samples by the scale factor. Consequently we have the following relation
Value of a sample - Value before AGC x Maximum limit
after AGC - Maximum value.

 This process is repeated for successive batches of samples until an appropriate total number of successive samples has been processed (e.g. 1000 samples). The multiplied samples are then stored in a memory.

 After the automatic AGC gain control step, the multiplied samples can be subjected to a series of checks or tests in order to detect an abnormal physiological condition. These checks can be carried out using a computer containing a program which includes a series of subroutines for the execution of the various checks. In the description which follows, details will be given of a series of checks which can be carried out to detect ventricular fibrillation.

 Such control depends on the presence of a substantial proportion of zero values (isoelectric segments) in the waveform of normal sinusoidal rhythm, while the proportion of zero values involved in ventricular fibrillation is very small. This is highlighted in fig. 3a which is an electrocardiogram corresponding to a normal sinusoidal rhythm, and in FIG.

4a which is an electrocardiogram corresponding to a ventricular fibrillation. If the computer performs a subroutine to measure the proportion of samples having a value of zero, a proportion greater than a predetermined degree can be considered as "normal n while a proportion less than this degree is considered to indicate possible fibrillation.

 As shown in fig. 3a, the baseline of a normal sinusoidal rhythm signal is not perfectly isoelectric due to noise and the presence of P and T waves, which may have passed the bandwidth filter in an attenuated state. This baseline irregularity can be avoided by using a floating zero, which can be set by the operator to an appropriate value, for example 20% of the maximum limit above the true baseline. Figures 3b and 4b-show the influence of this operation respectively on the curves of Figures 3a and 4a. We can see that the content of zero of the normal sinusoidal rhythm curve (NSR) is increased in a large proportion while that of ventricular fibrillation (VF), while being slightly increased, is still much lower than that of NSR.

 Using this test, we can consider as normal a proportion of zero value for at least 680 samples out of 1000 (68%).

 The validity of this test can be affected by a deformation created by the bandwidth filter. Some types of EKGs can be distorted to produce curve elements of the shape shown in fig. 5, and the presence of these elements decreases the content of zero of the curve. To eliminate this influence, it is possible to reverse the signal and repeat the test mentioned above, always using the floating zero. In most cases, a signal inversion increases the content by zero and the signal is again classified as a normal signal.

The distortion of the original signal which is caused by the filter can be eliminated by eliminating the filter and introducing another subroutine in the program to eliminate unwanted frequencies. However, the use of such a subroutine makes the program much more complicated and costs the computer time, which is detrimental in a system of this nature operating in
real time ".

 The test described above can also be ambiguous due to the existence of other possible cardiac conditions, for example ventricular tachycardia, which also correspond to a low content of zero but which are distinct from fibrillation. There is shown in FIG. 6 part of an electrocardiogram showing a ventricular tachycardia. A ventricular tachycardia corresponds to a high heart rate but it is distinguished from fibrillation in that the beats are essentially regular, so that the ratio between the energies contained in the electrocardiogram above and below the isoelectric line is essentially constant Consequently, if we measure the energies above and below the isoelectric line for an appropriate number of successive samples, and if we determine the ratio of said energies, by repeating this calculation for successive groups of samples, we can compare the energy ratios of successive groups or "niche fl.

If these ratios are essentially equal, then fibrillation can be eliminated as a possible condition.

 It has been found that the energy ratio established as indicated above is notably affected by the presence of baseline noise, which is random and does not want to be controlled. The effect of baseline noise can be eliminated by establishing amplitude limits above and below the baseline and by eliminating the signals between these limits of the total amounts of energy which are located above and below the isoelectric line. These limits are chosen experimentally.

Elimination of the noise effect achieved in this way decreases the possibility of variation of the energy ratios which are obtained from the normal sinusoidal rhythm.

 For each "slot", an appropriate duration of 2 seconds can be adopted, which corresponds to 500 successive electrocardiograph readings at 0.004 second intervals. It has been found that durations much shorter than this value (for example i second) do not allow very weak heart rhythms to be analyzed. Periods of 3 seconds or more tend to equalize energy ratios, even in the case of irregular heart activity, so that a "normal" result can be achieved when in fact fibrillation occurs.

 By using "slots" of 2 seconds and eliminating the noise as mentioned above, a result of this test is considered "normal" if the energy ratio is less than 20% of the average ratio for at least 7 out 10 slots.

 Another way of performing this test is to measure the energy ratios of successive slots, as described above, and calculate the variance. However, this process is less certain that a report drawn from a single niche, and which differs widely from the others, has a very strong influence on the variance.

 A third test to distinguish a normal sinus rhythm from a ventricular fibrillation is to statistically analyze the slopes of the output signal from the electrocardiograph. Figures 3a and 4a show that the NSR curve contains a high proportion of zero gradient and that most of the non-zero gradients are relatively steep. On the other hand, the VF curve contains little zero gradient and gives rise to a more or less random distribution of the non-zero gradients. If the curves are differentiated and a histogram showing the statistical distribution of the slopes of each curve is drawn, the histogram corresponding to the NSR curve has a much narrower distribution of slopes than the histogram of VF.

The width of these histograms can be expressed algebraically by the variance (the square of the standard deviation) of the population of the gradients of the curves. The variance of a curve can be calculated using a program subroutine, namely by differentiating the ECG electrocardiograph curve by appropriately processing the samples stored in the computer and by distributing the gradients obtained by a series of sorting boxes C (for example + 20, +10, +5, 0, -5, -10, -20). Variance # 2 is then calculated by the subroutine from the following equations Cfl x 1 f, X2 - --- fnx,
2 g 1 (xl x) + fa (X2 - x) 2 ... -n (xn - x) 23 where x = average value of the gradients obtained
N = number of gradients obtained
Xn = value of a gradient
fn = frequency of x
n = number of sorting boxes.

 A variance less than 75 was considered to represent a normal SNR heart rate, and a variance greater than this value indicates ventricular fibrillation.

 The above mentioned tests together make it possible to safely detect ventricular fibrillation but there are certain types of tachycardia for which they can indicate a false indication of fibrillation since the values of the content of zero and of the variance can come out outside the limits mentioned above. Certain types of tachycardia which can give rise to such an incorrect detection have been represented by the electrocardiograms of fig. 7a-d and they can meet at least two of the above criteria to indicate fibrillation.

 However, it should be noted that all curves 7a to 7d contain at least one weakened or deformed QRS complex occurring at regular intervals and that this complex is associated with the strongest negative slope of the entire activity cycle of the heart. . If this slope is detected and if the interval between successive detections the interval R-R) is measured, a comparison of the intervals indicates the presence or absence of a regular QRS complex.

 This comparison can be made by differentiating an appropriate number of successive samples (for example 250) and taking the strongest negative slope as a reference value. The rest of the stored samples are then differentiated and analyzed and the time intervals where similar slopes appear are recorded.

A similar tt slope "is a slope that falls within a predetermined range on either side of the reference value: an appropriate range is 20%.

 This analysis is continued until an appropriate number of similar slopes have been detected (for example 8). The intervals R-R existing between them are then compared and, if a predetermined proportion of said intervals (for example 5 out of 8) falls within predetermined limits, an absence of fibrillation is recorded.

If this criterion is not satisfied, an "irregular rhythm is indicated and, if no detection is made in a given period, for example of 3 seconds, the indication" no rhythm detection "is obtained.

 This test may give an incorrect result if an incident occurs during the "debugging" period when the reference value is established. Any incident risks producing a negative slope which is steeper than that of the QRS complex and which is therefore selected as the reference value. This possibility can be eliminated by adopting the process of establishing a reference value twice for adjacent batches of 250 samples and then taking the lower of the two values thus obtained as reference value.

 In the attached drawings, fig. 8 schematically shows a flow diagram of a pattern recognition system which is intended to fulfill the functions mentioned above.

 With reference to this diagram, a conventional electrocardiograph, comprising electrodes intended to receive electrical signals coming from a patient, applies the signals directly to a filtering unit 2. This unit 2 comprises a bandwidth filter "Sallen and Key" , whose circuit has been indicated in fig. 9 together with the values of the components of said circuit. The frequency response of this filter is approximately as shown in fig. 10.

 The filtered signal is then transmitted to an analog-digital converter of conventional type 3, where the signal is sampled at 0.004 second intervals and where the samples are expressed in digital form.

The digitally converted signals are then transmitted to a computer for further processing.

 In an appropriate arrangement, the analog-to-digital converter is an 8-bit converter which.

forms an interface with a mini-computer of the type
ALPHA 16, all the operations for processing the samples coming from the converter being carried out digitally by subroutines of the computer program.

 The digitally converted signals are first multiplied in the automatic gain control stage 4 by a scale factor in relation to the maximum value observed in 250 successive signals in order to establish calibrated values, as explained above. : the maximum limit value can be determined by an operator. The calibrated samples are stored in a memory and, when this memory contains 1000 successive samples, the samples are subjected to a series of tests which will be described below.

 The samples leaving stage 4 are applied to an energy ratio calculation stage 5 in which energy ratios for batches of 250 samples, representing periods of 2 seconds, are calculated as described above. The steps of this operation have been represented in flowchart 11. After setting the initial registers, the batches of samples are applied to them and they are compared with the positive and negative values, predefined by the operator and corresponding to limits d amplitude respectively placed above and below the isoelectric line (step 5a). Samples greater than this positive value are memorized and added together for the batch of 500 successive samples (step 5b) while samples less than the negative value are similarly stored and added together (step 5c). The total obtained after step 5b is then divided by that obtained after step 5c (step 5d) and the ratio is itself stored (Se). This process is repeated for successive batches of samples, so as to obtain a series of energy reports, until the samples obtained for 12 successive seconds of the signal from the electrocardiograph have been processed (step 7). We thus obtained a series of 6 energy reports, derived from 6 successive batches of 500 samples each.

 The samples from the automatic gain control step 4 are also transmitted to a zero content and slope variance 6 step, which has been shown in more detail in diagram 6. After adjusting the necessary initial registers, the samples from step 4 are applied to them. The samples are then compared with a floating zero established by the operator, for example a value 20% greater than the baseline of the maximum limit (step 6a). In step 6b, it is determined If each sample is greater than or less than the floating zero: if a sample is greater than the floating zero, this sample is memorized with its value whereas, if it is less, it is the value of the floating zero which is memorized (steps 6c and 6d). In step 6e, this process is continued until a batch of 1000 samples has been processed and memorized in this way.

 The successive samples of the batch are then differentiated with respect to time, using a differentiation factor of 7 C step 6f) so as to produce a series of gradients from successive groups of stored samples. The gradients obtained from groups of samples from step 6d, which are in fact samples defined with respect to the floating zero, are zero; the electrocardiograph samples from step 6c give non-zero gradients. All these gradients are treated in step 6g where the proportion of the zero gradients involved in the total number of gradients obtained is calculated. The proportion of null gradients is alor # s # emorIsed.

 The gradients obtained are also treated in step 6h, where the gradients are classified in a series of sorting boxes having respective values of 20, 10, 5, O, -5, -10 and -20. The totalization of the number of gradients in each box is then carried out (step 6i) and the statistical variance of the frequency distribution thus obtained is calculated (step 6j). The variance value obtained for the batch of 1000 samples is then memorized.

 When an analysis has been made over 12 seconds of the electrocardiograph signal, as described above, the stored quantities thus obtained are compared with predetermined criteria. The proportion of zero gradients obtained in step 6g is compared with a limit value which is set at 68% C step 8) and, if this proportion exceeds this value, the electrocardiograph signal is classified as normal using this criterion; if the proportion is lower, an "abnormal" signal is generated.

The variance value obtained in step 6j is compared with the limit value of 75 (step 9) and, if the variance is less than this value of 75, the electrocardiograph signal is classified as normal using this criterion on the contrary if the value is greater, an "abnormal" signal is generated.

The average value of the energy reports from step 5d is calculated (step 9a) and the number of reports which is less than 20% of this average is calculated (step 10). The electrocardiograph signal is classified as normal by this criterion if at least 4 out of 6 of the energy ratios fall within said limits; if at least 3 reports go beyond said limits, a signal
abnormal "is generated.

 If any 2 of these 3 criteria indicate a normal EKG signal, that is, if only one or no "abnormal" signal is generated (step 11), an output signal indicating an absence of fibrillation is issued (step 12). However, if at least 2 of said criteria indicate an abnormal electrocardiograph signal, a rhythm detection routine 13 is initiated to preserve the presence of fibrillation.

 The rhythm detection routine has been represented in flowchart 13. After adjusting the necessary initial registers, the samples from the automatic gain control 4 are memorized, 250 successive samples are differentiated and the strongest negative slope for the gradients obtained is determined (step 13a) so as to provide a reference value Limits of + + 20% of this reference value are then established (13b). The rest of the samples are differentiated (13c) and, in step 13d, the gradients are analyzed to detect values falling within the limits determined in step 13b. If no gradient falling within said limits has been detected during a period of 3 seconds (ie for 750 samples), step 13d causes the generation of a signal "no rhythm detection.

If a gradient falling within the above limits has been detected, a counter 13e is started to measure the time elapsing before the detection of the next slope falling within said limits. If no slope is detected in an interval of 3 seconds following the first detected slope, a signal "absence of rhythm detection" is again emitted (step 13f); if the elapsed time is less than 3 seconds, the recorded time is memorized (step 13g).

The counter is then reset to zero (step 13h) and the process corresponding to steps 13e to 13h is repeated until a total of 8 time values between gradients has been accumulated.

 The average of these time values is then calculated (13i) and limits of + 20% of this average value are established - (13d). The number of individual times falling within said limits is then defined (13k) and, if at least 5 of the 8 times are included in said limits, a normal signal n "is emitted in step 12. If 4 or less of said times go beyond said limits, a signal W detection of ventricular fibrillation "is emitted.

 The rhythm detection routine therefore indicates fibrillation when the maximum negative gradient is not repeated for 3 seconds or when at least half of the intervals between said gradients differ from the average value by more than 20%.

 To avoid an erroneous result due to an incident, steps 13a and 13b can be repeated for successive batches of 250 samples and the lesser of the values of the reference gradients thus obtained is taken as the reference value.

 The system described above is capable of analyzing an electrocardiograph signal and of indicating a possible fibrillation in less than 15 seconds and it can be used in a hospital service, in particular a high surveillance service, to quickly indicate a bad functioning of a patient's heart. Instead of or in addition to the generation of a visual or acoustic signal, the device can be connected to a defibrillator for the administration of a defibrillation substance to the patient during the indication of fibrillation.

The process described above for the detection of ventricular fibrillation can be implemented using a computer of known type which contains a program allowing the execution of the necessary logic operations. The computer can be connected to a graphic display terminal used to visually display the variable signal from the patient when a fibrilation indication signal is sent by the computer and to make a permanent recording of the signal from the patient if this is necessary. is necessary. The computer may comprise one or more microprocessors, suitably programmed from among the microprocessors which can be used, there may be mentioned those which are commercially available under the following designations: Motorola 6800, Intel 8080, 8085, 8086 and 8058, Zilog Z80 and Z8000, Mostck / Fairchild 3870,
Texas Instruments 9900 and RCA 1802 series. Using a microprocessor can provide a completely portable unit capable of detecting fibrillation and administering a defibrillation substance for use in emergency situations. on site as well as in hospitals.

Claims (24)

1. A shape recognition apparatus, characterized in that it comprises two electrodes arranged to be connected to the body of a patient so as to receive an electrical signal varying with respect to an isoelectric line, means for sampling the signal received at intervals and to convert the magnitudes of the samples from an analog form to a digital form, automatic gain control means for calibrating the magnitudes of the digitally converted samples with respect to a predetermined maximum value and a computer arranged to analyze the calibrated samples to detect an abnormal physiological condition. 2. Apparatus according to claim 1, characterized in that the computer is arranged to carry out the total of the quantities of the samples respectively greater and less than said isoelectric line for successive batches of samples, to calculate the ratios of the respective totals thus obtained , to compare the reports of successive batches and to issue a signal indicating an abnormal condition if the proportion of reports falling within predetermined limits situated above and below the average of the reports is less than a predetermined value. 3. Apparatus according to claim 2, characterized in that the computer is arranged to establish limits situated respectively above and below the isoelectric line and in that only the samples placed above and below the upper and lower limits respective are subject to a totalization. 4. Apparatus according to one of claims 2 or 3, characterized in that the computer is arranged to calculate the average of the reports for successive lots, to calculate the proportion of reports falling within the limits corresponding to + 20% of the average value and to generate a signal indicating an abnormal condition if said proportion is less than a predetermined value. 5. Apparatus according to claim 4, characterized in that the average of 6 successive batches is calculated and in that the signal indicating an abnormal cardiac condition is emitted if said proportion is less than 4 to 6. 6. Apparatus according to any one of claims 2 to 5, characterized in that each batch contains the samples covering a period of 2 seconds. 7. Apparatus according to claim 1, characterized in that the computer is arranged to calculate the proportion of samples of zero magnitude from a batch of successive samples and to generate a signal indicating an abnormal condition if said proportion is less than a predetermined value. 8. Apparatus according to claim 7, characterized in that the computer is arranged to establish a limit placed above the isoelectric line and to count all the sizes of samples falling below this limit considered as zero. 9. Apparatus according to claim 8, characterized in that the computer is arranged to invert the samples, to calculate again the proportion of samples of zero magnitude and to generate a signal indicating an abnormal condition if said proportion is less than a predetermined value both for inverted samples and for non-inverted samples. 10. Apparatus according to one of claims 8 or 9, characterized in that the computer is arranged to establish a limit located above the isoelectric line and corresponding to 20% of the maximum size of the calibrated digital samples, and to generate a signal indicating an abnormal condition if less than 68% of the sample sizes are counted as zero. 11. Apparatus according to claim 1, characterized in that the computer is arranged to differentiate groups of samples from a batch as a function of time in order to determine the gradient of each group, to calculate the variance of the gradients in the batch and to generate a signal indicating an abnormal condition if the variance exceeds a predetermined value.  12. Apparatus according to claim 1, characterized in  e that the computer is arranged to differentiate a group  successive samples of a batch as a function of time, for  calculate the maximum negative slope obtained, for different  the following groups of the batch, to calculate the inter  time intervals separating successive negative slopes  which fall within predetermined limits which are  respectively located above and below said  maximum negative slope, taken as a reference value,  for to calculate the average of said time intervals, for  calculate the proportion of intervals falling within  predetermined time limits above and below  below said average and to generate an indi  as for an abnormal condition if said proportion is  less than a predetermined value.  13. Apparatus according to claim 12, characterized in  that said predetermined slope limits are respec  20% higher and lower than the negative slope  maximum.  14. 14. Apparatus according to one of claims 12- or 13,  characterized in that said predetermined limits of  times are respectively 20% higher and lower than  said average.  15. Apparatus according to any of the claims  12, 13, 14, characterized in that 8 successive intervals  are calculated and in that a signal indicating a condition  abnormal is generated if no more than 4 intervals enter  within said predetermined time limits.  16. Apparatus according to any one of claims 12 to 15, characterized in that a signal indicating a condition  abnormal is generated if one of said time intervals  exceeds 3 seconds.  17. Apparatus according to any one of claims  12 to 16, characterized in that two groups of samples  successive of a batch are different, in that the slope  maximum negative for each group is calculated and in this  that the lower of the two slopes obtained is taken as  reference value. 18. Apparatus according to claim 1, characterized in that the computer is arranged to carry out the total of the quantities of the samples respectively located above and below the isoelectric line for successive batches of samples, to calculate the ratios of the totals thus obtained, to calculate the average of the ratios, to calculate the proportion of the ratios which fall within predetermined limits situated above and below the average and to generate a first signal if this proportion is less than a predetermined value, in that the computer is further arranged to calculate the proportion of samples of zero magnitude in a batch of successive samples and to generate a second signal if this proportion is less than a predetermined value, in that the computer is furthermore arranged to differentiate the groups of samples of a batch as a function of time in order to determine the gradients of each group, to calculate the variance of the gradients in the batch and to generate a third signal if the variance exceeds a predetermined value, in that the computer is further arranged to generate a fourth signal If two of said first, second and third signals are generated, and what the computer is further arranged to perform, in response to this fourth signal, the differentiation of a group of successive samples of a batch as a function of time, the calculation of the maximum negative slope obtained, the differentiation of following groups of the batch, calculating the time intervals separating successive negative slopes which fall within predetermined limits of slopes which are respectively greater and less than said maximum negative slope, calculating the average of said time intervals, calculating the proportion of the intervals falling within the predetermined time limits respectively above and below said average and the generation of a fifth signal indicating an abnormal condition if the said proportion is less than a predetermined value. 19. Apparatus according to any one of claims 1 to 18, characterized in that it comprises a filter placed between the electrodes and the signal sampling means in order to eliminate frequencies from the signals which are placed above and in below corresponding upper and lower limits. 20. Apparatus according to claim 18, characterized in that the filter is arranged to eliminate frequencies below 3 Hz and above 18 Hz. 21. Apparatus according to any one of claims 1 to 20, characterized in that it comprises a visual or acoustic warning device which can be actuated by the warning signal. 22. Apparatus according to any one of claims 1 to 21, characterized in that it comprises a defibrillator arranged to administer a defibrillation substance to the body of a patient in response to the signal indicating an abnormal condition. 23. A pattern recognition method, characterized in that a variable electrical signal is received from a patient and is sampled at intervals, in that the quantities of the samples are converted from an analog form to a digital form and are calibrated against a predetermined maximum value and that the calibrated samples are analyzed for an abnormal physiological condition.  24. A method of diagnosing ventricular fibrillation, characterized in that an electrical signal received from electrodes attached to a patient is sampled at intervals, in that the sizes of the samples are converted from an analog form to a numerical form and are calibrated with respect to a predetermined maximum value and in that the calibrated samples are analyzed in order to detect signal parameters characterizing ventricular fibrillation.